The key to Imagen Video's abilities is a "cascade" of seven diffusion models that transform the initial text prompt (such as "a bear washing the dishes") into a low-resolution video (16 frames ...